Description

Standalone voice server that turns text into speech. Any application sends text over HTTP or CLI — nayru cleans it, splits it into natural sentences, synthesizes audio through Kokoro TTS, and plays it back natively with gapless sequencing. Built as infrastructure for voice-enabled applications where the caller shouldn't have to think about audio pipelines.

Stack / Tools

• Rust
• Kokoros (Rust-native Kokoro TTS, ONNX, OpenAI-compatible API)
• rodio (audio playback)
• HTTP API / CLI
• Native audio (ALSA/PulseAudio/PipeWire, CoreAudio)

How it works

You send text, nayru handles the rest. Under the hood:

1. Clean & split — Markdown is stripped (code blocks, bold, headings, links, etc.) and the text is broken into individual sentences.
2. Synthesize — Each sentence is sent to a local Kokoro TTS server. Two workers run in parallel: while one streams the current sentence's audio to speakers, the other is already fetching the next sentence. This means the first audio plays in under a second, and subsequent sentences are ready by the time the previous one finishes.
3. Play — Raw PCM audio streams directly into a native audio sink (rodio) for gapless playback. No files, no buffering the whole response.

Stopping is instant — a single atomic counter invalidates all in-flight work.

Features

• Text-to-speech pipeline with markdown stripping and sentence splitting
• Kokoro TTS backend (local, OpenAI-compatible /v1/audio/speech)
• PCM streaming with pipelined prefetch (~0.3-0.9s time-to-first-audio)
• Audio queue with gapless playback via rodio
• HTTP server (default port 2003) with permissive CORS
• CLI: speak, stop, skip, pause, resume, status
• Embeddable as a Rust library
• Configurable voice, speed, and Kokoro URL

Instructions

1. Install and run Kokoros on port 3001:

   git clone https://github.com/lucasjinreal/Kokoros.git
   cd Kokoros
   bash download_all.sh
   cargo build --release
   ./target/release/koko --instances 1 openai --port 3001

2. Ensure an audio output device is available (ALSA/PulseAudio/PipeWire on Linux, CoreAudio on macOS)
3. Build with cargo build --release (binary at target/release/nayru)
4. Start the server with nayru serve or use the CLI / HTTP API

License

MIT

Usage

Server mode

# Start the voice server (default: 127.0.0.1:2003)
nayru serve

# With options
nayru serve --port 2003 --voice af_heart --kokoro-url http://localhost:3001 --speed 1.0

Client commands

# Send text to speak
nayru speak "Hello, this is nayru."

# Control playback
nayru stop      # Stop all speech, clear queue
nayru skip      # Skip current clip
nayru pause     # Pause playback
nayru resume    # Resume playback
nayru status    # Get current state

HTTP API

All endpoints served on port 2003 with permissive CORS.

Endpoint	Method	Body	Response
/speak	POST	{"text": "...", "voice": "af_heart"}	{"ok": true, "queued_chunks": 3}
/stop	POST	—	{"ok": true}
/skip	POST	—	{"ok": true}
/pause	POST	—	{"ok": true}
/resume	POST	—	{"ok": true}
/status	GET	—	{"state": "playing", "queue_length": 2, "voice": "af_heart"}

curl -X POST localhost:2003/speak -H 'Content-Type: application/json' -d '{"text":"Hello from curl"}'
curl localhost:2003/status

As a library

use nayru::tts::{TtsEngine, TtsConfig};

let engine = TtsEngine::new(TtsConfig::default());
engine.speak("Hello world.");
engine.status();  // TtsStatus { state: Playing, queue_length: 0, voice: "af_heart" }
engine.stop();

Building

cargo build --release

The binary is at target/release/nayru.